Skip to first unread message
S. MEIER
Apr 4, 2019, 12:46:15 PM4/4/19
to lavaan
Hello everyone,
I am still very new to lavaan and SEM and hope you can help me with a problem I can't seem to solve.
I have a mediation model with two dependent variables. I use a second order mediator. The basic model works just fine and shows good fit when I run it. Then I introduced some control variables, which are all binary variables and work in a similar previous model I defined in which, however, I did not use a second order form of the mediator, but the underlying latent dimensions directly. The model still seems just fine with the second order structure for some of the controls, but as soon as I introduce a few others, it shows NA for some of the fit measures. I am not sure what this is due to, it doesn't change when I take out some of the regressors on the left hand side of the regression equations, so it must be the controls themselves.
Its the upper two lines within the control section of the model that won't work.
Please find my model specification and output below. Thanks very much in advance!!
med.secord.mod <- '
# Latent Variables
GA =~ cca_01 + cca_02 + cca_03 + cca_04 + cca_05 #+ cca_06 + cca_07
IA =~ cca_08 + cca_09 + cca_10 + cca_11
WA =~ cca_12 + cca_13 + cca_14
CCA =~ GA + IA + WA
# Latent Variables
GA =~ cca_01 + cca_02 + cca_03 + cca_04 + cca_05 #+ cca_06 + cca_07
IA =~ cca_08 + cca_09 + cca_10 + cca_11
WA =~ cca_12 + cca_13 + cca_14
CCA =~ GA + IA + WA
cca_03 ~~ cca_04
cca_01 ~~ cca_05
cca_01 ~~ cca_05
JS =~ js_01 + js_02 + js_04
js_01 ~~ js_02 + js_04
js_01 ~~ js_02 + js_04
#Controls
CCA + JS + pri_total ~ v_77_3 + v_77_4 +
CCA + JS + pri_total ~ v_79_2 + v_79_3 + v_79_4
CCA + JS + pri_total ~ v_78_new + v_80 + v_108 + v_81_new
CCA + JS + pri_total ~ EX + AGR + CON + NEU + OP
CCA + JS + pri_total ~ v_77_3 + v_77_4 +
CCA + JS + pri_total ~ v_79_2 + v_79_3 + v_79_4
CCA + JS + pri_total ~ v_78_new + v_80 + v_108 + v_81_new
CCA + JS + pri_total ~ EX + AGR + CON + NEU + OP
#Regressions
CCA ~ a*cds
JS ~ b_j*CCA + c_j*cds
pri_total ~ b_p*CCA + c_p*cds
pri_total ~~ JS
indirect1 := a*b_j
indirect2 := a*b_p
direct1 := c_j
direct2 := c_p
total1 := indirect1 + direct1
total2 := indirect2 + direct2'
lavaan 0.6-3 ended normally after 163 iterations
Optimization method NLMINB
Number of free parameters 86
Number of observations 156
Estimator ML
Model Fit Test Statistic 397.094
Degrees of freedom 290
P-value (Chi-square) 0.000
User model versus baseline model:
Comparative Fit Index (CFI) NA
Tucker-Lewis Index (TLI) NA
Loglikelihood and Information Criteria:
Loglikelihood user model (H0) 57019.265
Loglikelihood unrestricted model (H1) -2319.008
Number of free parameters 86
Akaike (AIC) -113866.530
Bayesian (BIC) -113604.242
Sample-size adjusted Bayesian (BIC) -113876.459
Root Mean Square Error of Approximation:
RMSEA 0.049
90 Percent Confidence Interval 0.036 0.060
P-value RMSEA <= 0.05 0.565
Standardized Root Mean Square Residual:
SRMR 0.059
Parameter Estimates:
Information Expected
Information saturated (h1) model Structured
Standard Errors Standard
Latent Variables:
Estimate Std.Err z-value P(>|z|)
GA =~
cca_01 0.719 0.064 11.276 0.000
cca_02 0.673 0.060 11.206 0.000
cca_03 0.500 0.064 7.823 0.000
cca_04 0.552 0.063 8.796 0.000
cca_05 0.569 0.068 8.310 0.000
IA =~
cca_08 0.751 0.064 11.747 0.000
cca_09 0.619 0.067 9.200 0.000
cca_10 0.718 0.065 11.110 0.000
cca_11 0.670 0.066 10.177 0.000
WA =~
cca_12 0.415 0.080 5.162 0.000
cca_13 0.387 0.076 5.102 0.000
cca_14 0.382 0.075 5.088 0.000
CCA =~
GA 0.580 0.105 5.531 0.000
IA 0.451 0.097 4.630 0.000
WA 1.526 0.394 3.875 0.000
JS =~
js_01 0.135 0.193 0.701 0.483
js_02 0.173 0.247 0.701 0.484
js_04 0.142 0.201 0.708 0.479
Regressions:
Estimate Std.Err z-value P(>|z|)
CCA ~
v_77_3 0.279 0.310 0.900 0.368
v_77_4 0.234 0.361 0.647 0.517
JS ~
v_77_3 -0.778 1.611 -0.483 0.629
v_77_4 -1.012 1.966 -0.515 0.606
pri_total ~
v_77_3 -0.157 0.233 -0.675 0.500
v_77_4 -0.354 0.271 -1.306 0.191
CCA ~
v_79_2 -0.503 0.494 -1.017 0.309
v_79_3 -0.078 0.357 -0.220 0.826
v_79_4 0.388 0.578 0.671 0.502
JS ~
v_79_2 3.471 5.399 0.643 0.520
v_79_3 0.965 1.858 0.519 0.604
v_79_4 0.219 1.986 0.110 0.912
pri_total ~
v_79_2 -0.180 0.372 -0.486 0.627
v_79_3 0.155 0.267 0.580 0.562
v_79_4 0.074 0.433 0.172 0.864
CCA ~
v_78_new -0.354 0.343 -1.031 0.302
v_80 0.080 0.069 1.168 0.243
v_108 0.094 0.096 0.977 0.328
v_81_new 0.517 0.243 2.128 0.033
JS ~
v_78_new 1.932 3.111 0.621 0.535
v_80 -0.587 0.896 -0.655 0.513
v_108 -0.388 0.673 -0.577 0.564
v_81_new 0.022 0.860 0.026 0.980
pri_total ~
v_78_new -0.299 0.258 -1.160 0.246
v_80 0.090 0.052 1.739 0.082
v_108 0.036 0.072 0.499 0.618
v_81_new 0.218 0.185 1.180 0.238
CCA ~
EX -0.042 0.070 -0.595 0.552
AGR 0.048 0.081 0.587 0.557
CON 0.233 0.091 2.571 0.010
NEU -0.090 0.071 -1.257 0.209
OP 0.093 0.077 1.207 0.228
JS ~
EX 0.038 0.251 0.151 0.880
AGR -0.384 0.629 -0.611 0.541
CON -0.106 0.397 -0.267 0.789
NEU -0.466 0.675 -0.690 0.490
OP -0.396 0.654 -0.606 0.545
pri_total ~
EX 0.003 0.053 0.056 0.955
AGR 0.035 0.061 0.574 0.566
CON -0.085 0.069 -1.227 0.220
NEU 0.018 0.054 0.336 0.737
OP 0.051 0.058 0.880 0.379
CCA ~
cds (a) 0.001 0.016 0.046 0.963
JS ~
CCA (b_j) 4.109 6.088 0.675 0.500
cds (c_j) 0.063 0.105 0.598 0.550
pri_total ~
CCA (b_p) -0.262 0.093 -2.813 0.005
cds (c_p) -0.001 0.012 -0.045 0.964
Covariances:
Estimate Std.Err z-value P(>|z|)
.cca_03 ~~
.cca_04 0.228 0.056 4.076 0.000
.cca_01 ~~
.cca_05 -0.230 0.049 -4.696 0.000
.js_01 ~~
.js_02 0.074 0.052 1.412 0.158
.js_04 0.224 0.060 3.726 0.000
.JS ~~
.pri_total 0.053 0.363 0.147 0.883
Variances:
Estimate Std.Err z-value P(>|z|)
.cca_01 0.243 0.059 4.151 0.000
.cca_02 0.338 0.053 6.394 0.000
.cca_03 0.631 0.076 8.323 0.000
.cca_04 0.552 0.068 8.076 0.000
.cca_05 0.524 0.077 6.805 0.000
.cca_08 0.277 0.051 5.403 0.000
.cca_09 0.507 0.067 7.588 0.000
.cca_10 0.338 0.054 6.233 0.000
.cca_11 0.422 0.060 7.056 0.000
.cca_12 0.285 0.051 5.573 0.000
.cca_13 0.379 0.056 6.707 0.000
.cca_14 0.392 0.057 6.830 0.000
.js_01 0.593 0.085 6.999 0.000
.js_02 0.340 0.074 4.597 0.000
.js_04 0.551 0.076 7.292 0.000
.pri_total 0.831 0.098 8.433 0.000
GA 1.000
IA 1.000
WA 1.000
.CCA 1.000
.JS 1.000
Defined Parameters:
Estimate Std.Err z-value P(>|z|)
indirect1 0.003 0.067 0.046 0.964
indirect2 -0.000 0.004 -0.046 0.963
direct1 0.063 0.105 0.598 0.550
direct2 -0.001 0.012 -0.045 0.964
total1 0.066 0.115 0.574 0.566
total2 -0.001 0.012 -0.060 0.952
Terrence Jorgensen
Apr 5, 2019, 5:33:29 AM4/5/19
to lavaan
however, I did not use a second order form of the mediator, but the underlying latent dimensions directly.
What do you mean? You used each indicator as a distinct mediator instead of the common factor? What is "higher order"?
Your output says ML estimation. Any categorical endogenous variable must be declared using the ordered= argument, to trigger DWLS estimation
(thus fitting a probit regression to model the latent response underlying the discrete observed response).
(thus fitting a probit regression to model the latent response underlying the discrete observed response).
Terrence D. Jorgensen
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
S. MEIER
Apr 5, 2019, 2:03:13 PM4/5/19
to lavaan
Dear Mr. Terrence,
thank you very much for the quick reply!
What I meant with regard to the underlying dimensions is that in a previous model, I did not create a second order factor for the latent construct "CCA" but instead used the three first order factors "GA", "IA" and "WA". I then specified regressions for each dimension individually. In theory, however, these dimensions all pertain to a common concept denoted "CCA". Thus, in order to reduce model complexity, I decided to attempt using a second order factor, which would be conceptually more correct. In the previous (first-factor) model, the controls work just fine, but now when I introduce them, the fit turns NA.
GA =~ cca_01 + cca_02 + cca_03 + cca_04 + cca_05 #+ cca_06 + cca_07
IA =~ cca_08 + cca_09 + cca_10 + cca_11
WA =~ cca_12 + cca_13 + cca_14
CCA =~ GA + IA + WA
IA =~ cca_08 + cca_09 + cca_10 + cca_11
WA =~ cca_12 + cca_13 + cca_14
CCA =~ GA + IA + WA
Thank you for your suggestion, I will try whether that works.
Best regards,
Sophie
Reply all
Reply to author
Forward
0 new messages